Compared with conventional psychological models, which use basic math equations, Centaur did a much better job of predicting behavior. Accurate predictions of how humans respond in psychology experiments are beneficial in and of themselves: For instance, scientists could use Centaur to pilot their experiments on a pc before recruiting, and paying, human participants. Of their paper, nevertheless, the researchers propose that Centaur may very well be greater than only a prediction machine. By interrogating the mechanisms that allow Centaur to effectively replicate human behavior, they argue, scientists could develop latest theories in regards to the inner workings of the mind.
But some psychologists doubt whether Centaur can tell us much in regards to the mind in any respect. Sure, it’s higher than conventional psychological models at predicting how humans behave—but it surely also has a billion times more parameters. And simply because a model behaves like a human on the skin doesn’t mean that it functions like one on the within. Olivia Guest, an assistant professor of computational cognitive science at Radboud University within the Netherlands, compares Centaur to a calculator, which might effectively predict the response a math whiz will give when asked so as to add two numbers. “I don’t know what you’ll find out about human addition by studying a calculator,” she says.
Even when Centaur does capture something vital about human psychology, scientists may struggle to extract any insight from the model’s hundreds of thousands of neurons. Though AI researchers are working hard to work out how large language models work, they’ve barely managed to crack open the black box. Understanding an infinite neural-network model of the human mind may not prove much easier than understanding the thing itself.
One alternative approach is to go small. The second of the 2 studies focuses on minuscule neural networks—some containing only a single neuron—that nevertheless can predict behavior in mice, rats, monkeys, and even humans. Since the networks are so small, it’s possible to trace the activity of every individual neuron and use that data to work out how the network is producing its behavioral predictions. And while there’s no guarantee that these models function just like the brains they were trained to mimic, they’ll, on the very least, generate testable hypotheses about human and animal cognition.
There’s a price to comprehensibility. Unlike Centaur, which was trained to mimic human behavior in dozens of various tasks, each tiny network can only predict behavior in a single specific task. One network, for instance, is specialized for making predictions about how people select amongst different slot machines. “If the behavior is actually complex, you would like a big network,” says Marcelo Mattar, an assistant professor of psychology and neural science at Recent York University who led the tiny-network study and likewise contributed to Centaur. “The compromise, after all, is that now understanding it is vitally, very difficult.”
This trade-off between prediction and understanding is a key feature of neural-network-driven science. (I also occur to be writing a book about it.) Studies like Mattar’s are making some progress toward closing that gap—as tiny as his networks are, they’ll predict behavior more accurately than traditional psychological models. So is the research into LLM interpretability happening at places like Anthropic. For now, nevertheless, our understanding of complex systems—from humans to climate systems to proteins—is lagging farther and farther behind our ability to make predictions about them.
